Robot grabbing detection based on lightweight YOLOv10 algorithm

Traditional industrial robots face challenges such as low detection accuracy and low efficiency in target grabbing, which restricts the efficient development of industrial intelligent manufacturing. Therefore, the research proposes a robot grabbing detection technology based on lightweight target detection algorithm. This technology is based on the 10th generation target detection algorithm. It replaces the traditional backbone network by introducing a fast network, uses context guidance and dynamic upsampling modules to strengthen feature extraction and fusion, and uses a loss function to optimize the capture target position. In addition, the research also proposes a robot grasping posture optimization framework based on a generative grasping convolutional network, which optimizes the robot's grasping posture through weighted feature fusion and optimal grasping posture screening. In the comprehensive experimental test, the average precision, recall, and 50% intersection over union ratio threshold were 0.955, 0.971 and 0.942. In multi-scenario detection, the research model could detect 100% of target objects in dim scenarios and occlusion scenarios, and the detection accuracy was better than similar model performance. In the grasping posture optimization experiment, the grasping posture optimized model had a significant improvement in grasping detection accuracy compared to the unoptimized model. For example, in scenario 1, the average accuracy of the 50% intersection-to-union ratio threshold increased from 0.850 to 0.921, which significantly improved the robot grasping detection accuracy. The technology proposed in the study has good application effects, providing technical support for high-precision grabbing by industrial robots.